Submit a model
Add your TTS model for evaluation - public or anonymous.
Want your model on the board? Open an issue on GitHub, join our Discord, or email us and include an API endpoint and key (see the Provider API).
Submitting a model means accepting the Provider Agreement. The integrity policy is the short version - read that first at a minimum.
What we need
- An HTTP endpoint that synthesizes a line of text and returns audio.
- A pool of voices to rotate through (we don't do zero-shot cloning in the arena - we cycle a fixed set).
- A display name for the leaderboard.
- A model that's yours. Built or fine-tuned by you, or one you hold the rights to serve under your own name. We don't list wrappers, resellers, or rebrands of another provider's model - if the audio is coming from someone else's API, that provider is who belongs on the board.
Anonymous pre-release
You can run under a codename before your model is public - a way to get honest, blind feedback ahead of launch. Three rules keep this fair:
- Anonymous to voters, not to us. To run under a codename you tell the TTS Arena team who you are. We keep that confidential until the reveal - unless the integrity policy is broken, in which case we may disclose the provider as part of the outcome.
- Anonymity is time/exposure-bounded, not performance-bounded. A codenamed entry is revealed after a set period or vote count regardless of how it does - you can withdraw it before then, but you can't sit on the board indefinitely or reveal it only because it won.
- Permanent anonymity is for genuinely unreleased models. A model that's already publicly available runs under its real name (a codename for a fixed window is fine).
To keep comparisons fair, we don't evaluate multiple versions of the same model simultaneously.
Stealth models
Anonymous entries appear on the leaderboard under their codename with a neutral mark. Clicking one explains it's a stealth model in evaluation - its identity stays hidden until it's revealed.